> From: "Kastenholz, Frank" <FKastenholz@unispherenetworks.com>
> As to network mobility (where, loosely, we can assume a 1:1
> relationship between networks and IP prefixes), it seems to me that if
> the network moves, then its prefix "appears someplace else" in the
> topology.
Ah, I think perhaps not.
Either the name (i.e. your "prefix") says where it is in the network, in which
case you have to change it when you move, or it doesn't say where it is. In
the latter case, either the routing tracks the name wherever it goes (leading
to the scaling problem I just mentioned), or it doesn't - in which case it's
not an address, and you have to go through a translation step to get the
address.
If by a "network" you mean to include something like the collection of
computers inside a circa-2005 car, or something like that, you're talking
about many millions of mobile networks - and having the routing keep track of
millions of mobile networks wherever they go is no more plausible than having
the routing keep track of millions of mobile hosts wherever they go.
> Now there might be a whole lot of ways to deal with it, maybe as a part
> of the routing protocols, maybe as a separate protocol, maybe as
> something I haven't thought of.
Having said that, you get into a large swamp of potential solutions, about
which it's hard to say much.
E.g. if the mobility of the mobile network is limited in topological scope
(e.g. it can only plug into a limited part of the connectivity mesh), it may
be possible to do the mobility in the routing. (I.e. since the mobility is
limited in scope, the cost of supporting the mobility in the routing may
similarly be limited.) However, that restriction might not be feasible; e.g.
imagine a car rolling down a highway, connecting up to different wireless
ISP's as it goes along.
If you don't want to live with that restictions (limited network mobility),
then I would tend to think that perhaps mobile networks will need support from
the overall architecture (e.g. in the form of a "moblile network" object which
can group a number of endpoints together, among other properties).
Or perhaps the host mobility mechanism is good enough that they all rely on
that.
My crystal ball has clouded over.
Noel
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:04 EDT